Correlation Analysis

We can examine the relationship between search interest and mortality:

Correlations between Search Interest and COVID-19 Deaths by Theme and Search Term
Theme Search Term (Spanish) Search Term (English) Correlation P-value Significance
Socioeconomic Desempleo Covid Covid unemployment 0.785 0.000 ***
Prevention Mascarilla Face masks (alt) 0.685 0.000 ***
Prevention Sana distancia Social distancing 0.584 0.000 ***
Prevention Cubrebocas Face masks 0.571 0.000 ***
Symptoms Fiebre Covid Covid fever 0.413 0.000 ***
Basic Information Síntomas Covid Covid symptoms 0.393 0.001 ***
Medical Care Hospital Covid Covid hospital 0.359 0.002 **
Government Response Cuarentena Quarantine 0.355 0.003 **
Vaccination Efectos vacuna Vaccine side effects -0.338 0.004 **
Symptoms Pérdida olfato Loss of smell 0.322 0.007 **
Basic Information Covid Covid 0.321 0.007 **
Government Response Restricciones Covid Covid restrictions -0.277 0.020
Vaccination Vacuna Covid Covid vaccine -0.274 0.022
Government Response Semáforo Covid Covid traffic light system -0.228 0.058 ns
Socioeconomic Apoyo Covid Covid support 0.225 0.061 ns
Symptoms Pérdida gusto Loss of taste 0.211 0.079 ns
Basic Information Covid-19 Covid-19 0.207 0.085 ns
Symptoms Oxígeno sangre Blood oxygen -0.204 0.090 ns
Vaccination Registro vacuna Vaccine registration -0.185 0.126 ns
Basic Information Covid19 Covid19 0.155 0.200 ns
Prevention Quédate en casa Stay at home 0.142 0.240 ns
Symptoms Saturación oxígeno Oxygen saturation 0.137 0.257 ns
Basic Information Coronavirus Coronavirus 0.137 0.258 ns
Medical Care Oxígeno Oxygen -0.117 0.335 ns
Basic Information Prueba Covid Covid test -0.076 0.533 ns
Prevention Gel antibacterial Hand sanitizer 0.025 0.837 ns
Basic Information Test Covid Covid test (alt) -0.005 0.969 ns
Correlations between Theme-Aggregated Search Interest and COVID-19 Deaths
Theme Number of Terms Correlation P-value Significance SD of Interest
Prevention 5 0.504 0.000 *** 6.956
Socioeconomic 2 0.369 0.002 ** 1.078
Government Response 3 0.345 0.003 ** 7.307
Symptoms 5 0.287 0.016
0.295
Vaccination 3 -0.260 0.030
7.218
Basic Information 8 0.145 0.232 ns 4.525
Medical Care 4 -0.014 0.907 ns 1.578

Interpretation

The analysis reveals several interesting patterns in the relationship between Google searches and COVID-19 mortality in Mexico:

  1. Individual Search Terms (based on weekly correlations):
    • The strongest positive correlation was found for “Covid unemployment” (r = 0.785, p < 0.001), suggesting a strong relationship between economic concerns and mortality patterns
    • Face mask-related searches (“Mascarilla” and “Cubrebocas”) both showed strong positive correlations (r > 0.5), indicating consistent public interest in basic prevention measures
    • 13 out of 27 search terms showed statistically significant correlations (p < 0.05)
  2. Theme-Level Analysis:
    • Prevention-related searches showed the strongest theme-level correlation (r = 0.504, p < 0.001), suggesting that preventive measures were closely aligned with mortality trends
    • 2 themes showed negative correlations:
      • Vaccination (r = -0.260)
      • Medical Care (r = -0.014)
  3. Temporal Patterns:
    • Vaccination-related searches showed negative correlations (r = -0.260, p < 0.05), likely reflecting the timing of vaccination campaigns as mortality rates were declining
    • Socioeconomic searches maintained moderate positive correlation (r = 0.369, p < 0.01), suggesting persistent economic concerns throughout the pandemic
  4. Search Behavior Categories:
    • High-correlation themes (|r| > 0.4):
      • Prevention (r = 0.504)
    • Basic information searches showed weak correlation (r = 0.145, ns), suggesting consistent baseline information seeking regardless of mortality rates

Limitations

  • Analysis covers 27 search terms grouped into 7 themes over the period from December 2019 to May 2021
  • Google Trends data represents only internet users, which might not be representative of the entire Mexican population
  • Search patterns might be influenced by media coverage and public health campaigns rather than direct experience with COVID-19
  • The relative nature of Google Trends data (0-100 scale) makes absolute comparisons between terms difficult
  • Search behavior might differ across regions within Mexico, while our mortality data is national
  • Correlation doesn’t imply causation; many other factors influence both search behavior and mortality rates
  • The timing of searches might reflect both anticipatory behavior (searching before getting sick) and reactive behavior (searching after infection)

Cross-Correlation Analysis

Cross-correlation analysis helps to understand the temporal relationships between two time series. In our context, it helps us investigate whether and how search behaviors are related to COVID-19 mortality across different time shifts (lags).

What is Cross-Correlation?

Why Use Cross-Correlation?

Interpretation

The cross-correlation analysis reveals important temporal relationships between search behaviors and COVID-19 deaths, with different themes showing distinct patterns:

  1. Government Response
    • Shows strongest correlation at negative lags (around -5 weeks)
    • This suggests that government response-related searches preceded death counts by about 5 weeks
    • The declining pattern indicates that searches about government measures were more predictive than reactive
  2. Prevention
    • Exhibits consistently high correlation across lags
    • Slight peak at negative lags suggests preventive searches slightly preceded mortality
    • The sustained correlation indicates ongoing public interest in prevention throughout mortality waves
  3. Socioeconomic
    • Peak correlation at lag 0 to -1 weeks
    • Shows symmetric decline in both directions
    • Suggests concurrent relationship between economic concerns and mortality
  4. Symptoms
    • Shows strongest correlation at positive lags (around +5 weeks)
    • Indicates that symptom searches followed increases in death counts
    • May reflect growing public awareness after mortality waves
  5. Medical Care
    • Shows weak but increasing correlation at positive lags
    • Suggests medical care searches slightly lagged behind death counts
    • The pattern might reflect reactive healthcare information seeking
  6. Basic Information
    • Shows weak, relatively constant correlation across lags
    • Suggests consistent baseline information seeking
    • Little temporal relationship with mortality patterns
  7. Vaccination
    • Shows consistently negative correlation
    • Flat pattern across lags
    • Reflects the inverse relationship with mortality, likely due to vaccination timeline
  • Predictive Patterns: Government response and prevention searches tended to precede mortality increases
  • Reactive Patterns: Symptom and medical care searches tended to follow mortality increases
  • Concurrent Patterns: Socioeconomic searches showed strongest correlation with current mortality
  • Independent Patterns: Vaccination and basic information searches showed little temporal relationship with mortality

Seasonal Decomposition

To better understand the temporal patterns in COVID-19 related searches, we perform a time series decomposition and impact analysis focusing on the initial pandemic period.

Example for seasonal decomposition of COVID-19 search queries:

Changepoint Analysis of Search Patterns

Changepoint analysis identifies significant shifts in the mean level of search interest. We use: - PELT (Pruned Exact Linear Time) algorithm - Modified Bayesian Information Criterion (MBIC) for penalty selection - Focus on changes in mean level

Basic Information

Number of changepoints: 9

Period Mean Level
2019-12-29 to 2020-02-16 0.69
2020-02-16 to 2020-03-08 3.46
2020-03-08 to 2020-04-12 14.70
2020-04-12 to 2020-08-09 9.49
2020-08-09 to 2020-11-29 6.54
2020-11-29 to 2020-12-13 11.94
2020-12-13 to 2021-01-24 16.90
2021-01-24 to 2021-02-07 11.00
2021-02-07 to 2021-05-30 6.35
2021-05-30 to 2021-05-30 0.00

Government Response

Number of changepoints: 10

Period Mean Level
2019-12-29 to 2020-03-08 0.76
2020-03-08 to 2020-03-22 26.83
2020-03-22 to 2020-04-05 22.50
2020-04-05 to 2020-04-12 33.33
2020-04-12 to 2020-04-19 23.67
2020-04-19 to 2020-05-10 19.00
2020-05-10 to 2020-05-24 14.00
2020-05-24 to 2020-06-07 9.67
2020-06-07 to 2020-07-26 6.29
2020-07-26 to 2021-05-30 3.45
2021-05-30 to 2021-05-30 0.00

Medical Care

Number of changepoints: 4

Period Mean Level
2019-12-29 to 2020-04-12 0.64
2020-04-12 to 2020-12-13 2.24
2020-12-13 to 2021-01-31 6.04
2021-01-31 to 2021-05-30 2.07
2021-05-30 to 2021-05-30 0.00

Prevention

Number of changepoints: 10

Period Mean Level
2019-12-29 to 2020-02-16 2.83
2020-02-16 to 2020-03-08 10.73
2020-03-08 to 2020-03-15 35.20
2020-03-15 to 2020-04-05 23.53
2020-04-05 to 2020-04-19 28.60
2020-04-19 to 2020-04-26 21.20
2020-04-26 to 2020-08-02 16.87
2020-08-02 to 2020-09-06 11.44
2020-09-06 to 2021-02-28 8.08
2021-02-28 to 2021-05-30 5.35
2021-05-30 to 2021-05-30 0.00

Socioeconomic

Number of changepoints: 3

Period Mean Level
2019-12-29 to 2020-03-22 0.12
2020-03-22 to 2020-05-17 3.38
2020-05-17 to 2021-05-30 1.09
2021-05-30 to 2021-05-30 0.00

Symptoms

Number of changepoints: 1

Period Mean Level
2019-12-29 to 2021-05-30 0.35
2021-05-30 to 2021-05-30 0.00

Vaccination

Number of changepoints: 10

Period Mean Level
2019-12-29 to 2020-07-05 0.38
2020-07-05 to 2020-11-29 1.84
2020-11-29 to 2021-01-24 5.08
2021-01-24 to 2021-01-31 27.33
2021-01-31 to 2021-03-21 11.10
2021-03-21 to 2021-04-04 19.67
2021-04-04 to 2021-04-18 14.67
2021-04-18 to 2021-04-25 35.33
2021-04-25 to 2021-05-16 26.11
2021-05-16 to 2021-05-30 46.50
2021-05-30 to 2021-05-30 0.00

Summary of Search Interest Transitions

The changepoint analysis reveals distinct phases in public attention across different themes during the COVID-19 pandemic in Mexico (December 2019 - May 2021):

Key Phases and Transitions

  1. Initial Outbreak Phase (February - April 2020)
    • Sharp increases across multiple themes
    • Notable spikes in:
      • Government Response (peak: 33)
      • Prevention measures (peak: 35)
      • Basic Information (significant increase in March)
  2. First Wave Response (April - August 2020)
    • Gradual decline in government and prevention-related searches
    • Sustained moderate interest in basic information
    • Emergence of socioeconomic concerns (peak: 3.4)
  3. Second Wave (November 2020 - January 2021)
    • Renewed interest in basic information (peak: 17)
    • Increased medical care searches (peak: 6)
    • Initial vaccination interest emerging
  4. Vaccination Period (January - May 2021)
    • Most dynamic phase for vaccination searches
    • Multiple changepoints with increasing interest (final peak: 46)
    • Declining interest in other themes
  • Most Variable: Government Response and Prevention and Vaccination and Basic Information and Medical Care and Socioeconomic and Symptoms (10 changepoints each)
  • Most Stable: Symptoms (only 1 changepoint)
  • Most Recent Peak: Vaccination (May 2021)
  • Early Peak & Decline: Government Response and Prevention
  1. Search Evolution
    • Shift from prevention/response to vaccination focus
    • Decreasing interest in basic information over time
    • Short-lived peaks in government response measures
  2. Theme Interactions
    • Medical care interest aligned with outbreak waves
    • Socioeconomic concerns showed distinct early peak
    • Prevention measures tracked with government responses
  3. Public Attention Spans
    • Initial high interest followed by adaptation
    • Renewed attention during second wave
    • Sustained vaccination interest with multiple peaks

Impact Analysis of first lockdown (2020-03-01)

Impact analysis helps us understand how significant events during the pandemic affected public information-seeking behavior. Using interrupted time series analysis, we examine whether specific interventions or events caused substantial changes in search patterns.

Impact analysis, in this context, uses segmented regression to:

The analysis provides several key metrics:

Impact Analysis of March 2020 Intervention by Theme
Theme Immediate Effect Monthly Trend Change R²
Government Response 1149.3% *** -2.8% ns 0.802
Socioeconomic 279.7% *** -4.2% ns 0.653
Basic Information 237.9% *** -20.2% *** 0.889
Prevention 214.1% *** -13.8% *** 0.882
Medical Care 110.8% *** -8.2% ** 0.863
Symptoms 32.3% ** -2.6% ns 0.819
Vaccination -19.0% ns 7.0% ns 0.945
Significance levels: ** p<0.001, ** p<0.01, * p<0.05, ns: not significant
† Immediate Effect: Percentage change in search interest immediately after March 1, 2020
‡ Monthly Trend Change: Percentage change in monthly trend after intervention
§ R²: Proportion of variance explained by the model

Interpretation of Lockdown Impact Analysis

The implementation of COVID-19 measures in March 2020 had varying effects across different search themes:

Immediate Effects

  1. Strongest Responses (>200% increase):
    • Government Response showed the most dramatic surge (1149.3%)
    • Socioeconomic (279.7%) and Basic Information (237.9%) searches also increased substantially
    • Prevention-related searches rose significantly (214.1%)
  2. Moderate Responses:
  • Medical Care searches +110.8%
  • Symptoms searches +32.3%
  1. No Significant Impact:
  • Vaccination searches showed no significant immediate change (-19.0%)

Trend Changes

Significant Declining Trends:

  • Basic Information showed -20.2% monthly decline
  • Prevention showed -13.8% monthly decline
  • Medical Care showed -8.2% monthly decline
  • Most other themes showed non-significant trend changes
  • Only Vaccination showed a positive trend (+7.0%, though not significant)

Model Reliability

  • All models show strong fit (R² > 0.65)
  • Vaccination model shows highest explanatory power (R² = 0.945)
  • Basic Information and Prevention models also show very good fit (R² > 0.882)

This pattern suggests an initial surge of public interest in government measures and practical information, followed by a gradual decline in attention, possibly indicating public adaptation to the new situation.